Question Answering on ∞-Bench Longbook QA English (test)
11.2F1 ScoreNestedKV
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NestedKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 11.2 | — | — | |
| NestedKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 11.2 | — | — | |
| KeyDiffBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 10.98 | — | — | |
| PyramidKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 10.76 | — | — | |
| NestedKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 10.58 | — | — | |
| PyramidKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 10.4 | — | — | |
| PyramidKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 10.19 | — | — | |
| SnapKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 10.13 | — | — | |
| ExpAttnBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 10 | — | — | |
| KeyDiffBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 9.63 | — | — | |
| SnapKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 9.47 | — | — | |
| ExpAttnBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 9.06 | — | — | |
| SnapKVBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 7.77 | — | — | |
| KeyDiffBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 7.3 | — | — | |
| ExpAttnBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 6.79 | — | — | |
| StreamingLLMBackbone=Qwen3-4B, KV eviction ratio (r)=0.25, Input cap=40k2026.05 | 5.4 | — | — | |
| StreamingLLMBackbone=Qwen3-4B, KV eviction ratio (r)=0.50, Input cap=40k2026.05 | 5.11 | — | — | |
| StreamingLLMBackbone=Qwen3-4B, KV eviction ratio (r)=0.75, Input cap=40k2026.05 | 4.81 | — | — | |
| InfLLMContext Window=10242026.03 | — | 1,024 | 16.27 | |
| InfLLMContext Window=20482026.03 | — | 2,048 | 18.38 | |
| InfLLMContext Window=40962026.03 | — | 4,096 | 22.09 | |
| TokenSelectContext Window=10242026.03 | — | 1,024 | 13.71 | |
| TokenSelectContext Window=20482026.03 | — | 2,048 | 16.93 | |
| TokenSelectContext Window=40962026.03 | — | 4,096 | 18.44 | |
| UT-ACAMaximum Block Budget=8, Update=Sub.42026.03 | — | 896 | 16.58 | |
| UT-ACAMaximum Block Budget=16, Update=Sub.82026.03 | — | 1,664 | 18.72 | |
| UT-ACAMaximum Block Budget=32, Update=Sub.162026.03 | — | 3,712 | 21.32 |